Horizon Entropy

Ted Jacobson
Department of Physics, University of Maryland
College Park, Maryland 20742-4111

Renaud Parentani
Laboratoire de Mathématiques et Physique Théorique, UMR 6083
Université de Tours, 37200, France
arXiv:gr-qc/0302099v1 25 Feb 2003

Abstract
Although the laws of thermodynamics are well established for black hole horizons,
much less has been said in the literature to support the extension of these laws to
more general settings such as an asymptotic de Sitter horizon or a Rindler horizon (the
event horizon of an asymptotic uniformly accelerated observer). In the present paper we
review the results that have been previously established and argue that the laws of black
hole thermodynamics, as well as their underlying statistical mechanical content, extend
quite generally to what we call here “causal horizons”. The root of this generalization
is the local notion of horizon entropy density.

1 INTRODUCTION
Black hole thermodynamics was born a little over three decades ago from two simple
questions: “What is the most efficient way to extract the rotational energy of a black
hole?” and “Can the second law of thermodynamics be violated by dumping entropy
across a black hole horizon?” The answer to the first question is that the most efficient
energy extraction occurs when the horizon area remains constant, and that moreover
if the area increases the process is classically irreversible[1, 2]. More generally, Hawk-
ing’s area theorem[3] showed that the horizon area cannot decrease provided that only
positive energy can flow across the horizon.
Jacob Bekenstein’s proposed answer[4] to the second question was that it is indeed
not possible to violate the second law, provided one attributes to the horizon itself an
entropy proportional to its area. The analogy between horizon area and entropy was
already motivated by the energy extraction results and the area theorem. Bekenstein
added information theoretic arguments, interpreting the horizon area as a measure of
the missing information for an outside observer, and deducing that the entropy must
be a constant of order unity times the area in Planck units. He thus introduced the
generalized second law (GSL) of thermodynamics, which states that the sum of the
black hole entropy plus the matter entropy outside the black hole “never decreases”[5].
The GSL took on a more precise form after Hawking’s discovery[6] that a black hole
of surface gravity κ radiates at a temperature TH = κ/2π.1 When taken together with
the zeroth law[7] which states that the surface gravity is constant over the horizon, and
the first law[7] for isolated black holes,
κ
∆M = ∆A + Ω∆J (1)
8π
(where M and J are the mass and angular momentum of the black hole and Ω is the
angular velocity), this yielded the coefficient (1/4) of proportionality between horizon
1
We use in this paper Planck units, with ~ = c = G = k = 1.

1
area and entropy (as well as attributing a truly thermal character to a black hole). The
GSL is then the statement that no physical process can lead to a decrease of the sum of
the entropy outside a horizon Soutside and the Bekenstein-Hawking entropy SBH = A/4
of the event horizon,
∆(Soutside + A/4) ≥ 0 . (2)
The GSL was initially postulated with a positive energy condition in mind, in which
case the horizon area of a black hole can only increase. However, it turns out that the
GSL holds even during quantum evaporation of the black hole via Hawking radiation[8,
9], when a negative energy flux across the horizon produces a decrease of area. In
that case, the reason for the validity of Eq. (2) is that Hawking radiation carries
enough entropy so that the increase of Soutside outweighs the decrease of the black hole
entropy. Convincing arguments have been given which establish more generally that,
in any quasi-stationary process, the GSL will hold provided the ordinary second law
holds for matter[10, 11, 12]. Taken together, these results imply that black holes are
thermodynamic systems.
The generalization of thermodynamics from black hole horizons to de Sitter hori-
zons was initiated by Gibbons and Hawking[13, 14] shortly after the discovery of the
Hawking effect, and is by now quite well accepted, despite the observer-dependence of
de Sitter horizons. Nevertheless the corresponding generalization to acceleration hori-
zons is less well accepted, especially with regard to the entropy they carry, although
they bear a close resemblance to de Sitter horizons. Bekenstein’s original information
theoretic concept of the nature of black hole entropy certainly applies to all horizons,
even highly non-stationary ones, as does the area increase theorem (see below). The
ultimate significance of black hole thermodynamics hangs on this issue of generality,
and we thus consider it an ideal topic for our contribution to this festschrift for Jacob.
In this article we take a broad look, both historical and fresh, at the question how
general are the laws of horizon thermodynamics. One of the key questions is whether
the local notion of horizon entropy density is a valid concept, or whether something
essentially global is involved. At the classical level this question concerns the interplay
of horizon area and the gravitational field equations. At the quantum level it concerns
the role of horizon area in determining the density of states factor in partition functions,
in irreversible processes, and in transition rates. The latter statistical role is the more
fundamental, since it is related to the underlying quantum states, and it gives rise to
the thermodynamic laws.
Using both the classical and quantum viewpoints, we shall argue that local entropy
density of a horizon is indeed a valid concept, and that consequently the laws of black
hole thermodynamics extend quite generally to any causal horizon. At the statistical
level a causal horizon should thus be conceived as an ensemble whose mean properties
are the well-known geometrical ones. This conception was pioneered by Bekenstein,
who emphasized from the outset the physical relevance of random exchanges which
might some times lead to a decrease of the area[15]. From this vantage point, the GSL
emerges as “a statistical law which becomes overwhelmingly probable in the limit of a
macroscopic system”[5].
After all this, do we understand horizon entropy?[16] We close with a discussion
of the question what does horizon entropy count, a question whose answer we believe,
despite some glimpses, remains shrouded in mystery.

2
 2 CAUSAL HORIZONS
By a causal horizon we mean here the boundary of the past of any timelike curve λ
of infinite proper length in the future direction. This definition was introduced by
Gibbons and Hawking[13] under the name “event horizon”, however we shall call it a
“causal horizon” to emphasize its generality, since today “event horizon” refers almost
exclusively to the case of a black hole event horizon. More generally, the boundary of
the past of any set of events is a kind of causal horizon. This more local notion has not
been used as much in the context of horizon thermodynamics as that associated with an
infinite future. Nevertheless it does appear to have some thermodynamic or statistical
role to play, as evidenced for example in [17, 18]. Four examples are given in Fig. 1.

i+ I+ p
I+

asymp. flat black hole de Sitter

I+ I+
p p

anti-de Sitter asymp. flat

Figure 1: Carter-Penrose diagrams of causal horizons in (conformal compactifications of) various

spherically symmetric spacetimes. Each generic point in one of the diagrams represents a 2-sphere
of symmetry, and the radial null lines make 45 degree angles with the vertical. The points on the
left hand edge of each diagram lie at the origin of spherical symmetry, hence they represent just a
single point. The points on the right hand edge of the de Sitter diagram lie at the antipode of the
3-sphere, hence they also represent a single point. Future null infinity is denoted by I + .

For a black hole in asymptotically flat spacetime, the event horizon can be defined as
the boundary of the past of all of I + , however an equivalent definition is the boundary of
the past of any any timelike curve that goes to future timelike infinity i+ . Equivalently,
the event horizon is the boundary of the past of i+ itself. Therefore, the black hole
horizon is defined with reference to the intrinsic asymptotic structure of the spacetime,

3
without referring to a particular class of observers.
An asymptotic de Sitter horizon is defined in a spacetime that is asymptotically de
Sitter in the future. It is defined by the boundary of the past of a timelike worldline
reaching a point p at I + (which is spacelike in asymptotically de Sitter spacetime),
or equivalently the boundary of the past of p. Since there are many different points
at I + there are many inequivalent asymptotic de Sitter horizons. An asymptotic de
Sitter horizon is thus said to be “observer-dependent”, in contrast to a black hole event
horizon.
Our other two examples are observer-dependent in the same sense. An asymptotic
anti-de Sitter horizon, defined in a spacetime that is asymptotically anti-de Sitter at
spatial infinity, is the boundary of the past of a timelike worldline reaching a point
p at I + (which is here timelike,) or equivalently the boundary of the past of p. An
asymptotic Rindler horizon (ARH), defined in a spacetime that is asymptotically flat,
is the boundary of the past of an accelerated worldline that goes to a point p on I +
(which is here null,) or equivalently the boundary of the past of p.
Let us bring out the nature of an ARH in more detail. The boundary of the past
of p is a cone in compactified Minkowski spacetime, however it is actually a plane in
Minkowski space. To show this we use coordinates (u, v, x, y) for which the line element
takes the form ds2 = dudv − dx2 − dy 2 . The point p could be the endpoint v → ∞ of
the light ray (u0 , v, x0 , y0 ). The points (u, v, x, y) to the past of (u0 , v0 , x0 , y0 ) are those
with u < u0 and (x0 − x)2 + (y0 − y)2 < (u0 − u)(v0 − v). In the limit v0 → ∞ this
includes all points with u < u0 . The ARH is the boundary of this region, which is the
null plane u = u0 . Thus in Minkowski spacetime an ARH is just what is usually meant
by a “Rindler horizon”. Note that the location of this horizon is independent of the
transverse coordinates x0 and y0 of the light ray. It is nevertheless observer dependent
since it depends on u0 . Note also that the horizon has infinite cross sectional area.
An asymptotically flat spacetime approaches Minkowski space near I + , so an ARH in
general will asymptotically approach a Rindler horizon.
A causal horizon whose generators coincide with some Killing flow is a Killing hori-
zon, i.e. a null hypersurface generated by a Killing flow. A bifurcate Killing horizon
(in four dimensions) is a pair of Killing horizons which intersect in a particular two di-
mensional spacelike cross section—called the bifurcation surface—on which the Killing
vector vanishes. Examples of these occur in Minkowski, de Sitter, Anti de Sitter, and
Schwarzschild spacetimes. In the Minkowski case the pair of planes u = 0 and v = 0
(in the above mentioned coordinates) comprise a bifurcate Killing horizon for which the
Killing field is the hyperbolic rotation generator −u∂u +v∂v , and the bifurcation surface
is the x-y plane at u = v = 0. This is the bifurcate Rindler or acceleration horizon, and
the Killing field generating it is called the boost Killing field. The “other sheet” of the
Schwarzschild horizon is called the past horizon since it is defined by the boundary of
the future of past null infinity (in the same asymptotic region whose future defines the
future horizon). Similarly in Minkowski spacetime for example the other sheet is the
past acceleration horizon. It should be emphasized that the various examples of Killing
horizons are all indistinguishable in a neighborhood of the bifurcation surface smaller
than the curvature scale, hence the acceleration horizon provides a universal template
for all of them.
While the notion of bifurcate Killing horizon is a global one, it can be localized
in two ways. Obviously, one can focus on a neighborhood of the bifurcation surface
or even just part of it. Somewhat less obviously, a neighborhood of a piece of a single
Killing horizon can be extended to a neighborhood of a bifurcate Killing horizon includ-
ing the bifurcation surface, provided the surface gravity is constant and nonvanishing
on the horizon[19]. The constancy of the surface gravity—the zeroth law of horizon

4
thermodynamics—can be derived either using the Einstein equation and the dominant
energy condition[7] or from the assumption that a neighborhood of the horizon is static
or stationary-axisymmetric with a t-φ reflection isometry[20, 21].
In [13] Gibbons and Hawking argue that in empty spacetimes satisfying the Ein-
stein equations with a cosmological constant (or with electromagnetic fields) stationary
causal horizons are necessarily stationary axisymmetric Killing horizons. Their argu-
ment follows the same lines as those that had been given previously for the special case
of asymptotically flat black hole horizons. (Although not stated explicitly in [13], it
seems the horizon cross sections are assumed to be compact in these arguments.) They
also point out that the proof in Ref. [7] of the zeroth law for such Killing horizons
extends immediately to the case of a cosmological constant. Alternatively, the zeroth
law follows without the use of field equations or energy conditions as discussed above.

3 THE SECOND LAW OF CAUSAL HORIZONS

Gibbons and Hawking[13] quote a “second law of event horizons” (whose proof is referred
to a forthcoming publication that seems never to have appeared) which states that the
area of causal horizons cannot decrease provided that (i) what the observer can see at
late times can be predicted from a spacelike surface (a cosmic censorship assumption),
and (ii) the energy-momentum tensor satisfies the strong energy condition. Similar
theorems were quoted by Penrose[22] under both the strong and null energy conditions,
and with several different formulations of the cosmic censorship assumption (though we
have not been able to find their actual proofs in print.) It should be noticed that all
these theorems concern the classical version of the second law in which only positive
energy fluxes are considered.
A simple but not entirely satisfactory proof that causal horizons satisfy the classical
area theorem can be given under the assumptions that (i) the null energy condition
holds, and (ii) the null geodesics generating the horizon are future complete, i.e. have
infinite affine length in the future. This proof, which is just Hawking’s original proof
of the black hole area theorem[3], works by contradiction. Suppose the congruence of
horizon generators were converging at some point p. Then the null energy condition
implies that this focusing cannot be reversed, and will therefore produce a caustic that
would be reached at finite affine parameter along the generator to the future of p. Such a
caustic is in essence an intersection of infinitesimally separated null horizon generators,
which is incompatible with the defining property of the horizon as the boundary of the
past of an infinite timelike curve. Hence the expansion of the null congruence of horizon
generators must be greater than or equal to zero everywhere. That is, the area of any
bundle of horizon generators is non-decreasing.
What is not entirely satisfactory about the above proof is the assumption that the
horizon generators are future complete. In the black hole case this assumption can
be replaced by the cosmic censorship assumption that no singularities (to the future
of a partial Cauchy surface) are visible from I + [23, 24]. (To exploit this assumption
it is necessary to use the device of deforming the horizon slightly outward to another
surface containing points that are causally connected to I + .) For a general causal
horizon similar proofs were long ago reported[13, 22] (but not published) under dif-
ferent formulations of the cosmic censorship assumption. Recently, the area theorem
for general “future horizons” has been reexamined and proved by Chruściel et al.[25],
without assuming piecewise smoothness of the horizon.
Besides such general results, area theorems for various special cases of horizons
other than standard black hole horizons have been established by several authors. A
theorem was proved by Davies[26] for the case of asymptotically de Sitter causal horizons

5
in Friedmann cosmologies with a cosmological constant, under the assumption that
the fluid satisfies the dominant energy condition. An area increase theorem for black
holes in asymptotically de Sitter spacetime was proved by Hayward et al. [27] and
Shiromizu et al.[28]. An explicit proof of the area increase theorem for the cosmological
event horizon (defined as the past Cauchy horizon of I + ) in asymptotically de Sitter
spacetimes containing a black hole was given by Maeda et al.[29].

4 THE FIRST LAW OF CAUSAL HORIZONS

In this section we consider several versions of the first law of black hole thermodynamics
and discuss how they extend to all causal horizons. We close the section with the
argument that the ultra-local version of this first law implies, in a precise sense, the
Einstein equation as a corresponding equation of state.

Stationary comparison version

Having shown that causal horizons in stationary spacetimes satisfy the zeroth and sec-
ond laws, Gibbons and Hawking went on to derive a version of the first law. In par-
ticular, for a pair of infinitesimally close spacetimes with both cosmological and black
hole horizons, minus the variation of the integral of Killing energy density on a spatial
surface bounded by the horizons is given by

− δEKilling = (κC /8π)δAC + (κH /8π)δAH + ΩH δJH . (3)

The Killing vector ξ a employed here is the one that generates the cosmological horizon.
The Killing energy density is defined by Tab ξ a nb , where nb is the unit normal to the
spatial surface, κC , κH , AC and AH are the surface gravities and areas of the cosmo-
logical and black hole horizons, JH is the angular momentum of the black hole (defined
by the integral of the curl of the rotational Killing field over the horizon), and ΩH is
the angular velocity of the black hole relative to the cosmological horizon. No angular
momentum term appears for the cosmological horizon since the Killing vector ξ a gen-
erates that horizon. Note that although this Killing field has no natural normalization,
that presents no difficulty since each term in Eq. (3) scales with the undetermined
normalization in the same manner.
This version of the first law is a generalization of Eq. (1), which arises when taking
the asymptotically flat space-time limit κC → 0. The generalization involves not the
variation of the ADM mass—which is not defined in a de Sitter background—but instead
the variation of the matter Killing energy. It can thus be used to describe variations
in the asymptotically flat case between two stationary configurations at fixed ADM
energy, in which the matter energy is redistributed, with some going into the black
hole. Formulated this way, the first law is more local, since no reference to what is
happening at infinity is required.

Physical process version

Both of the previously mentioned versions of the first law Eq. (1) and Eq. (3) involve
comparisons of two stationary solutions to the field equations. A yet more local version
of the first law emerges when one enquires into the dynamical process by which the
horizon area adjusts to a small flow of energy across the horizon. This relation, which
Wald[30] called the “physical process version” of the first law in the case of black hole

6
 horizons, was first established in [31] (see also [32] for a comprehensive review.) The
physical process version of the first law states that

δS = δEH /TH , (4)

R
where δS = δA/4 is one fourth the horizon area change, δEH = H Tab ξ a dΣb is the flux
of “energy” across the horizon, defined with respect to the horizon generating Killing
field ξ a of the stationary background spacetime which is being perturbed, and TH is
1/2π times the surface gravity κ = |∇a ξ| of the horizon generating Killing field.
To exhibit its local nature, let us review the derivation of the physical process first
law. We first discuss the approach using affine parametrization and then discuss the
approach using Killing parametrization. The expansion θ of the null congruence of
horizon generators measures the fractional rate of increase of the cross-sectional area
element with respect to an affine parameter λ. Hence the change in the area of a bundle
of generators over a finite range of affine parameter is
Z
∆A = θ d2A dλ, (5)
B

where the integral is over a patch B of the horizon consisting of the finite range of the
chosen generators. This integral can be related to the energy flux using the identity
θ = d(λθ)/dλ − λdθ/dλ, together with the Raychaudhuri equation for dθ/dλ and the
Einstein equation. The Raychaudhuri equation is
dθ 1 1
= − θ2 − σ 2 − Rab k a k b , (6)
dλ 2 2
where σ 2 = σab σ ab is the squared shear of the congruence and k a is the affine tangent
vector to the congruence, and the Einstein equation reads
1
Rab − Rgab = 8πTab . (7)
2
Eq. (5) thus yields
Z Z
1 2
∆A = [λθA]21 + (θ + σ 2 ) d2 A dλ, + 8πTab λk a dΣb . (8)
B 2 B

Eq. (8) is an identity satisfied by any null congruence in a spacetime satisfying the
Einstein equation. It becomes the first law when we specialize to a situation involving
a small change of a stationary background spacetime, with the congruence correspond-
ing to a small deformation of the Killing horizon. We now see how this happens by
examining each term in the equation.
If the small deformation is caused by some matter stress tensor at order ǫ, then the
metric perturbation and therefore θ and σ will be of order ǫ. Hence the term (θ2 + σ 2 )
can be neglected.2
If the upper limit of integration is chosen in a region where the horizon is again
stationary, then the net area change of a given bundle of generators must be finite,
hence θ must vanish faster than 1/λ, so the term (λθA)2 vanishes. To understand the
behavior of the lower limit it is helpful to relate the affine parameter λ to the Killing
2
Actually, neglect of this term is not justified if the generators develop caustics or if the integration region
extends to infinity, as in the case of an asymptotic Rindler horizon, as will be discussed below.

7
parameter v of the background stationary spacetime. In general the relation between
affine and Killing parameters on a Killing horizon is

λ = aeκv + b, (9)

where a and b are arbitrary constants. If we exploit the shift freedom of the affine
parameter to set b = 0, then as the Killing parameter v goes into the past, the affine
parameter λ goes to zero exponentially fast. The term (λθA)1 is already first order in
ǫ from θ, hence to first order we can use this relation between the background values
of λ and v to conclude that this term is exponentially suppressed provided the stretch
of (background) Killing time over which the first law is being applied is long compared
with κ−1 . This temporal restriction of the applicability of the first law corresponds to
an “equilibration time” (albeit backwards in time). Indeed, the expansion defined with
respect to Killing time also vanishes exponentially.
The last step is to relate λk a in the energy flux term to the Killing vector ξ a . Since
the energy is a first order perturbation, this relation can be evaluated in the strictly
stationary background on the Killing horizon. We thus have λk a = λ(dv/dλ)ξ a , where
v is the Killing parameter. From the relation (9) with b = 0 we have λ(dv/dλ) = κ−1 ,
hence
λk a = κ−1 ξ a . (10)
When substituted into Eq. (8) this yields the physical process form of the first law Eq.
(4) when the patch B is taken to be a piece of the horizon bounded in the past and the
future by two complete cross-sections.

Adiabatic version
In the preceding formulation of the first law for physical processes it was assumed only
that the neighborhood of the horizon is nearly stationary, so that the perturbation is
small. If one assumes further that it is adiabatic, then an even more local form of the
first law applies. As pointed out above, the horizon equilibration rate is given by κ,
hence adiabatic means slow with respect to that scale. To extract the adiabatic limit,
it is useful to re-express the Raychaudhuri equation (6) in terms of derivatives with
respect to the Killing parameter v rather than the affine parameter. This produces the
equation
dθ̂ 1 1
= κθ̂ − θ̂2 − σ̂ 2 − Rab k̂ a k̂ b , (11)
dv 2 2
where θ̂ = (dλ/dv)θ, σ̂ = (dλ/dv)σ, k̂ a = (dλ/dv)k a and κ = (d2 λ/dv 2 )/(dλ/dv) (cf.
Eq. (9)). For a slow perturbation we have θ̂ ≪ κ, σ̂ ≪ κ, and dθ̂/dv ≪ κθ̂, so both the
left hand side and the second order terms in the expansion and shear can be dropped,
which yields
θ̂ ≈ κ−1 Rab k̂ a k̂ b . (12)
When this is substituted in Eq. (5) and the Einstein equation is used, we see that,
when the evolution is adiabatic, the first law Eq. (4) holds over any patch and over any
stretch of time of the slowly evolving horizon.

Extension to all causal horizons

While the preceding derivations apply for black hole horizons, they are localized on the
horizon and make no reference to spatial infinity. Therefore it is tempting to apply them
to all causal horizons. Possible obstructions concern the normalization of the Killing

8
 vector, the possibly infinite horizon area, and convergence of the upper limit. Let us
consider these in turn.
First, the energy flux δEH and the temperature TH scale in the same way under a
change of normalization of the Killing field. Hence, as in Eq. (3), this normalization
cancels out so no choice is required. (In the stationary black hole case the horizon
generating Killing field is a linear combination of a time translation and a rotation,
ξ a = (∂/∂t)a + ΩH (∂/∂φ)a . The time translation Killing field (∂/∂t)a is normalized at
infinity, so the change of “energy” in Eq. (4) is expressed as δM − ΩH δJ.)
Second, the derivation of the first law can be applied to a bundle of generators
forming a finite subset of the horizon, so no infinite area need be confronted. Moreover,
the change of total area δA can be finite, even if the total area itself is infinite. This
change can be defined by considering first a finite bundle of horizon generators whose
cross-sectional area is then taken to infinity.3
Third, convergence of the upper limit (λθA)2 of Eq. (8) can be analyzed in each
bundle of generators. As explained above this upper limit vanishes if and only if the
cross sectional area of any finite bundle is finite at infinity. This corresponds to the
condition that the horizon is in equilibrium in the asymptotic future. An ARH satisfies
this condition, since it is just a Rindler horizon at infinity. (We have not checked
whether the horizon of a point at I + in AdS is asymptotically stationary.)
The physical process first law thus applies to all situations in which a causal
horizon goes through an evolution that is nearly stationary and settles down
in the future.

Asymptotic Rindler horizons

In the limit that the cosmological constant goes to zero, the stationary comparison
version (3) of the first law for cosmological horizons becomes the first law for Rindler
horizons. To illustrate this, we suppose a rest mass δm follows a Killing orbit on which
the norm of the Killing field is ξ. The Killing energy of this mass is ξ δm, so the
first law (3) implies that the disappearance of the mass√increases the horizon area by
δAC = (8πξ/κC ) δm. In de Sitter space we have κC /ξ = a2 + Λ, where a is the proper
acceleration of the Killing orbit. If the mass sits at the center where a = 0, then in
the limit as Λ goes to zero the area change diverges. If instead a 6= 0 then the ratio
κC /ξ approaches a and the area change is finite. It is interesting to note that, when Λ
is nonzero, κC /2πξ is equal to the local Unruh/de Sitter temperature Tlocal that would
be experienced by an accelerated mass in the Euclidean vacuum state[34]. (As Λ goes
to zero this becomes just the Unruh temperature for the accelerated mass and as a goes
to zero it becomes just the de Sitter temperature.) The disappearance of the mass thus
increases the horizon entropy by an amount equal to δm/Tlocal for all values of Λ and
a.
Unlike the stationary comparison version, the physical process version (4) of the
first law does not hold for an asymptotic Rindler horizon since the ‘nearly stationary’
condition cannot be satisfied. We now explain the reason for this in two ways, first from
the perspective of the ARH itself, and then by describing the ARH as the infinite mass
limit of a Schwarzschild horizon.
Suppose a planet in free fall wanders across the ARH. Although the boost energy
flux of the planet is finite, it can be shown that the net area change caused by the
passage of the planet is infinite. This is essentially because every null geodesic on the
3
See [33] for an example of computing the finite difference of acceleration horizon areas when comparing
two different spacetimes.

9
 ARH—no matter how far from the planet in the transverse direction— is eventually
focused backwards in time to a focal point on the line behind the planet, opposite to
the direction of the endpoint of the ARH. That is, all the horizon generators originate
on the line behind the planet. Hence an infinite area is added to the ARH so evidently
the first law (4) is not satisfied.
The reason the physical process first law is not satisfied is that the process is not
nearly stationary. The Raychaudhuri equation (6) of course applies, but the terms
neglected to arrive at the first law cannot be neglected. For example this is clear from
the fact that there are focal points, where the expansion goes to infinity and hence θ2
cannot be neglected.4 Moreover, since all generators originate on the line of caustics,
the ultimate area of the annulus of generators between asymptotic cylindrical radii ρ
and ρ + ∆ρ must grow from zero without any energy having crossed the congruence.
According to (8) all of this area must come from integrating the squared expansion and
shear. The shear will dominate (since it is the shear that generates the expansion in this
case[31]). For any bundle of generators this produces a finite final area, however when
integrated over an infinite cross section and infinite affine parameter range it produces
an infinite total area change. Thus, even though the squared shear is O(ǫ2 ) in the
perturbation, it has an infinite net effect.
Further insight is offered by viewing the ARH as the infinite mass limit of a Schwarzschild
horizon of mass M . The planet crossing the ARH can be modeled by a planet that falls
in freely from infinity and crosses the Schwarzschild horizon. This situation has been
analyzed in [35], where it was shown √ that the process is nearly stationary for the pur-
poses of the first law only if R ≫ mM , where R is the radius of the planet and m
is the mass.5 In the limit of M going to infinity, this requires that the planet have an
infinite radius. Hence, for any finite sized planet crossing, an ARH will not satisfy the
physical process version of the first law. The fact that the area increase is infinite can
also be understood using the infinite mass limit, since the planet comes in from infinity
with finite Killing energy m and the surface gravity κ = 1/4M goes to zero. Hence,
if the physical process form of the first law were to apply, the area change would be
given by δA ∼ (8π/κ)m = 32πmM , which diverges as M goes to infinity.6 Failure of
the nearly stationary condition would only add to this divergence.
It should be emphasized that although the nearly stationary physical process first
law (4) does not hold, the expression (8) for the horizon area change remains valid. This
should be considered an extension of the first law to non-equilibrium thermodynamics,
as discussed at the classical level in [31, 32] and at the quantum statistical level in
[36, 37]. It would be interesting to pursue this line of thought further.

Local Rindler horizons and the Einstein equation of state

We have just seen how the Einstein equation enforces the first law for near-stationary
transitions of causal horizons. To conclude this section let us show how this logic
can be turned around to derive the Einstein equation as an “equation of state” of
spacetime[17]. To render the argument entirely local we make use of the notion of a
4
A similar but simpler example of a non-nearly stationary process is the formation of a black hole by an
infalling thin spherical shell of mass M . The horizon grows in area from the focal point in the center where
it originates, and this area growth is not properly accounted for by the first law. However, if a second such
shell with mass ∆M ≪ M follows, the transition does obey the first law.
5
This condition can also be expressed by saying that the surface area of the planet is much larger than
the increase of the horizon area according to the first law.
6
As the planet falls, a small fraction (∼ 0.01m/M ) of the rest mass is radiated as gravitational waves,
hence the Killing energy crossing the horizon approaches m as m/M goes to zero.

10
“local Rindler horizon” (LRH). An LRH is defined as the boundary of the past (on one
side) of a spacelike two-surface element S through a point p, with S adjusted so that
the generators of the LRH have vanishing expansion and shear at p. Thus the LRH
reaches equilibrium at p. The area change of a small bundle of generators approaching
the “equilibrium point” p is given by Eq. (5), with

θ ≈ −λ Rab k a k b , (13)

where the affine parameter λ has been set to zero at p, and k a is the affinely parametrized
horizon generator. If we restrict attention to a sufficiently small neighborhood, then
the effect of any classical energy momentum tensor makes only a small perturbation of
the local Minkowski space, hence the near-stationary analysis holds. According to Eq.
(10) we can replace −λk a by κ−1 ξ a , where ξ a is the boost Killing field on the LRH.
The first law Eq. (4) will hold on the small bundle of generators, with TH equal to the
Unruh temperature associated with ξ a , if and only if at p we have

Rab k a k b = 8πTab k a k b . (14)

If this is to hold for all LRH’s through p then it must hold for all k a , which implies
that Gab = 8πTab + f gab , where f is an arbitrary function. The Bianchi identity and
the conservation of Tab then imply that f is a constant, which can be identified with an
undetermined cosmological constant.
It is difficult to resist concluding from this argument that the horizon entropy density
proportional to area is a more primitive concept than the classical Einstein equation,
which appears as a thermodynamic consequence of the interplay of entropy and causal-
ity. From the point of view of the quantum action principle this is conclusion is not so
surprising. The action gives rise to the entropy at the quantum statistical level, and to
the field equations at the level of the classical mean field. It is to the former, deeper
level of description that we now turn.

5 STATISTICAL MECHANICS OF CAUSAL HORIZONS

In this section we discuss, at the quantum statistical level, the extension of the notion
of black hole Boltzmann entropy to stationary states of all causal horizons, i.e. to
Killing horizons. The starting point for this generalization is the work of Gibbons
and Hawking[14] which showed that semiclassical evaluation of the partition function
for the gravitational field yields A/4 for the entropy of both black holes and de Sitter
space. Here we examine the role of horizon entropy in determining the rates for dynamic
processes, and we argue that this role extends in particular to acceleration horizons. We
also briefly consider a few approaches to explicit “state counting” of horizon degrees of
freedom.

Pair creation
A first indication in favor of the extension of the statistical notion of black hole en-
tropy to all causal horizons arises from the pair creation probability for magnetically
charged black holes in a magnetic field[38]. This probability is given as a product of
two exponentials:
P (q, M ; B) ∝ eA/4 × e∆Aaccel /4 . (15)
The second factor contains ∆Aaccel (q, M, B): the change of area of the acceleration
horizon[33] which is associated with the creation of the black hole pair. This term

11
 persists even when no black holes but only monopoles are created, and it reproduces
the well known instanton contribution exp(−πM 2 /qB) in the test field approximation
G → 0 where G is Newton’s constant. The first factor tells us that the probability is
enhanced[38, 33] over that of monopoles by exp(A/4), where A(q, M ) is the horizon
area of a black hole of mass M and magnetic charge q.7
This is consistent with the usual statistical interpretation of A/4 as the logarithm of
the number of black hole states. From Eq. (15) we see that the two area changes deter-
mine the probability in the same manner. It is therefore compelling to apply the same
statistical interpretation to both of them. Perhaps the strongest indication in favour
of treating both areas on the same footing arises from the fact that an accelerated
black hole will be in a statistical ensemble since it will experience Unruh radiation. The
random exchanges of thermal photons between the acceleration horizon and the hole
will bring the whole system into equilibrium[39]. It is to be noticed that the equilib-
rium situation is determined by Eq. (15) and corresponds to the maximum probability
where ∂M (A + ∆Aacc ) = 0, i.e. when Hawking and Unruh temperatures agree. In this
Ssystem = (A + ∆Aacc )/4 does act as the entropy of the whole system. The random
energy exchanges will also spread both the black hole mass and the area of the accelera-
tion horizon around their mean values. The spread in the mass of the hole is related to
2 −1/2
the heat capacity and it is given, as usual, by (∂M Ssystem )equil . We emphasize that the
conclusion that the acceleration horizon is in an ensemble requires no assumption other
than that the creation probability Eq. (15) also delivers the equilibrium distribution.
The considerations of the following subsection support this natural assumption.
Consideration of pair creation of black holes in de Sitter space yields again a prob-
ability given as the exponential of two area changes[40]. In this case, people would
likely agree that both terms possess a statistical interpretation. Despite the similarity
between the two cases, Hawking and Horowitz[41] nevertheless assert that the accel-
eration horizon area should not be interpreted as counting an entropy. They give two
reasons. The first is that the usual evaluation of the entropy from the partition function
cannot be performed since the period of the Euclidean section is fixed and cannot be
varied. The second is that the acceleration horizon is observer dependent and hence
the information behind the horizon “can be recovered by observers who simply stop
accelerating”. Restated, they assert the unitarity of the evolution on a foliation cutting
across the acceleration horizon. The last two statements can be made for a black hole
horizon (or a de Sitter horizon) however, so they do not seem to preclude the association
of area with entropy for the acceleration horizon.

Transition rates for systems in contact with a Killing horizon

Further evidence for the statistical entropy of causal horizons is obtained from [42] where
it is explained why the transition rates for systems in contact with a Killing horizon are
governed by changes in the horizon area. The analysis applies to black hole horizons
with the evaporation process itself, to acceleration horizons with the Unruh effect expe-
rienced by accelerated systems, and to cosmological horizons with the processes related
to the de Sitter temperature. In all cases, when taking into account the gravitational
back-reaction, the excitation rates are governed by exp(δA/4) in place of the thermal
Boltzmann factor exp(−δEK /TH ) which is known to govern Hawking radiation[43] for
7
One might have thought that it should be the total area of the two black holes appearing here, however
that is not the case. The reason is that the pair carries the quantum numbers of the vacuum, so their
quantum degrees of freedom are strictly correlated[39]. Thus the degeneracy of states is that of a single
black hole.

12
example. In these expressions, δEK is the change in Killing energy, TH the horizon
temperature measured in the same units as δEK , and δA is the corresponding area
change. In the test field approximation, δA is related to the matter change by Eq. (3).
Hence we recover the transition rate in a thermal state, as previously obtained in the
case of Hawking radiation or transitions of an accelerated detector[44]. The novelty
consists of taking into account the back-reaction of the spacetime geometry in the eval-
uation of individual transition amplitudes (like one does it with recoil effects[45]), and
no longer in the mean as one does when using the semi-classical Einstein equations. As
a byproduct, one notices that canonical distributions (governed by Hawking or Unruh
temperature) are replaced by microcanonical distributions in which the heat capacity
of the horizon is properly accounted for.
To understand how this replacement occurs, recall that the action must sometimes
be supplemented by boundary terms in order to have a well-defined action principle,
since the action should be stationary when varied within the class of metrics under
consideration. Consider Sh , the action for matter+gravity in its hamiltonian form and
applied to the class of metrics which belong to the “one black hole sector”[46], namely
metrics which are asymptotically flat and which also possess an inner horizon. The
total variation of Sh reads
Z
δA
δSh = dΣ3 (πδg + pδq)|21 + bulk terms + δM (t2 − t1 ) − (Θ2 − Θ1 ) . (16)
8π
The first terms give the initial and final momenta of gravity and matter respectively.
The bulk terms vanish on-shell. The third term arises from the outer boundary. It is
given by the variation of the ADM mass times the lapse of asymptotic proper time.
The last term arises from the inner boundary and it is given by the change of the
horizon area times the lapse of the hyperbolic angle Θ. (On-shell and for stationary
metrics, dΘ/dt = κ gives the surface gravity.) For the class of processes involving
exchanges between a black hole and surrounding matter at fixed ADM energy M R , the
outer boundary term vanishes. Hence one should not add to Sh the energy term − dtM .
On the other hand, since the area then adjusts itself dynamically, it should not be fixed.
Thus, one must get rid of the inner boundary term in Eq. (16) by working with the
(Legendre transformed) exchange action
A
Sexchange = Sh + Θ. (17)
8π
As noticed by Carlip and Teitelboim [46], quantization of this action yields for the
matter + gravity wave function the unusual Schrödinger equation

d Â
i Ψ = − Ψ, (18)
dΘ 8π
where  is the operator valued horizon area which acts on both the matter and gravity
sectors. The system at fixed ADM energy thus evolves according to Θ and not the time
at infinity. Eq. (18) is remarkable. It shows that the area itself is the operator-valued
geometrical quantity that replaces the Hamiltonian in governing exchanges.
Two applications of Eq. (18) were considered in [42]. First is the thermalisation of
a two level atom held at some fixed distance from a black hole. This type of situation
also applies to the thermalisation of inertial systems in de Sitter space and leads to
expressions that can be put in parallel to the static version of the first law Eq. (3).
The other case concerns self-gravitating effects within Hawking radiation as studied in
[47, 48]. The results found there emerge simply and transparently when the system

13
is analyzed using the conjugate variables Θ and A rather than the pair of asymptotic
variables t and M .
The essential new point is that, via the gravitational back reaction, each matter
energy eigenstate j is correlated to an area eigenstate with eigenvalue Aj . Thus it
evolves according to (18) with a Θ-time dependence given by exp(iAj Θ/8π). (The
entanglement of the gravitational subspace of eigenvalue Aj to the matter state j is not
specific to the present case: because of the constraints, this gravity-matter entanglement
always occurs when working with the solutions of the Wheeler-DeWitt equation[49].)
As an illustration the system can be taken to be a two-level atom held fixed at
some distance from the hole (for a more complete discussion see [42]). To mediate
the interactions we introduce a radiation field which allows for transitions from ground
state j to excited state j ′ by absorption or emission of a photon. To first order in the
coupling between the atom and the radiation field, the transition amplitude from j to
j ′ accompanied by the emission of a photon of frequency Ω is given by
Z
Tj→j ′ +Ω = C dΘ Ψj (Ψj ′ φΩ )∗
Z
−Θ
= C̃ dΘ ei(Aj −Aj′ )Θ/8π e−iΩe . (19)

In the first line, the wave function Ψj characterizes the time evolution of the initial
state of the system: a black hole plus the detector in its ground state with the radiation
field in its vacuum state. The product in parenthesis characterizes the final state:
the black hole and the excited detector plus the photon of frequency Ω. Their wave
functions factorize because we neglect the gravitational back-reaction of the radiation.
(This is a legitimate assumption when computing the rates to leading order in the back-
reaction.) The overall constants C and C̃ will play no role since we shall be interested in
the ratio R = |Tj→j ′ +Ω /Tj ′ →j+Ω |2 which determines the equilibrium distribution. In
the second line, we have grouped together the two ‘unperturbed’ stationary states with
their corresponding area-eigenvalues. The last factor is the wave function of a photon of
(Kruskal) frequency Ω. Its properties are fixed by the following considerations. Since we
are interested in near equilibrium transitions, the initial state of the radiation field must
be stationary and regular on the horizon. It must therefore be either the Hartle-Hawing
vacuum[43] or the Unruh vacuum, depending on the condition one imposes on infalling
configurations. It is most economical (though physically less direct) to characterize one-
particle states by making use of modes which individually incorporate the requirement
of regularity at the expense of being non-stationary. (The other option would be to work
with stationary modes (like Schwarzschild modes e−iλΘ ) and to express the Unruh or
Hartle-Hawking state as a thermal state in terms of these modes.) When working with
the first option, the modes are of the form exp(−iΩU ), where U is a null outgoing
coordinate which is regular on the future horizon (see e.g. [50] for a discussion of the
various relevant modes). When evaluated at the (stationary) location of the two level
atom, one has U (Θ) = −c e−Θ , where the constant c can be put to unity by a shift in
Θ.
Before proceeding, it is of interest to display the corresponding expression one would
obtain by working in a fixed background. Instead of Eq. (19) one would have
Z
Tj→j ′ ,Ω = C ′ dt ψj (ψj ′ φΩ )∗
Z
−κt
= C̃ dt e−i(Ej −Ej′ )t e−iΩe ,
′
(20)

14
where Ej and Ej ′ are the eigen-energies of the two atom states measured with the
asymptotic time which defines the surface gravity κ.
When working with the non-stationary modes, the ratio R of transition rates can
be simply evaluated, at fixed Ω, by recognizing from (19) that Tj ′ →j+Ω is related to
T∗j→j ′ +Ω by a shift Θ → Θ − iπ of the Θ contour. This yields R = exp(Aj ′ − Aj )/4
for the case when the gravity-matter entanglement has been taken into account in
the amplitudes. On the other hand when using the background fixed transitions the
ratio of the rates is R = exp −(Ej ′ − Ej )(2π/κ), which gives rise to a Boltzmann
distribution with the horizon temperature κ/2π. As mentioned above, the first law Eq.
(3) guarantees that these results agree to first order in G. The only difference is that,
in the first case, the specific heat of the horizon has been taken into account through
the gravity-matter entanglement of the wave functions. Let us emphasize that in both
cases, the ratio of the transition rates has been computed unconventionally by making
use of the (non-stationary) Kruskal modes. The stationarity of the processes follows
from the fact that the above ratios are independent of Ω.
The principle of detailed balance implies that the ratio pj ′ /pj of equilibrium occu-
pation probabilities for the detector + horizon system, with the field in the Unruh or
Hartle-Hawking state, is given by the ratio R = exp(Aj ′ − Aj )/4 of transition rates.
The equilibrium distribution is also the microcanonical distribution, hence we infer that
the number of states in a configuration with area A is proportional to exp A/4. Since
the above analysis is couched in terms of A and Θ, which are defined at the horizon, it
is entirely local to the neighborhood of the Killing horizon and therefore takes the same
form for black hole, de Sitter, and acceleration horizons. It was therefore concluded in
[42] that a quantum statistical interpretation of the entropy A/4 should apply to all
“event horizons”.
A difference between the Rindler horizon and both the black hole and de Sitter ones
is that in the former case the temperature of the horizon is perceived as a physical
temperature only by observers who follow the accelerated worldlines of the horizon
generating Killing field (i.e. it is the Unruh temperature), while in the black hole and de
Sitter cases there are also unaccelerated observers who perceive a nonzero temperature.
This difference plays no role in the above analysis however.

Other approaches to counting horizon states

Of the many other approaches to calculating black hole entropy from quantum first
principles, it seems to us that none present any evidence that the notion of universal
horizon entropy density is invalid. For examples we briefly consider here the approaches
of Carlip[51] using the structure of the near-horizon component of the diffeomorphism
group, the canonical quantum gravity calculations in the loop approach[52], and the
calculations of string theory.
Carlip’s approach, being based on representations of the residual local diffeomor-
phism symmetry modulo horizon boundary conditions, is as applicable to any Killing
horizon as it is to black hole horizons. It does not rely on the global properties of
the horizon, but instead appears consistent with the notion of horizon entropy den-
sity. The loop quantum gravity approach as currently formulated involves an induced
Chern-Simons theory on the horizon, and applies to both compact black hole and de
Sitter horizons. For technical reasons however it does not apply to non-compact Rindler
horizons or even to a Rindler horizon compactified to a 2-torus. It is not yet clear what
result would be obtained for the entropy of a Rindler horizon using this approach.
The string calculations are of two types. In the weakly coupled D-brane context,
near-extremal configurations at weak coupling are found to have the same entropy as

15
the black hole with the corresponding charges at strong coupling (for a review see [53]).
To apply this reasoning to other causal horizons seems problematic, since it is unclear
how to identify an appropriate weakly coupled cousin. The second (and related) string
approach is in the context of the AdS/CFT duality, wherein the black hole entropy is
identified with ordinary thermal entropy in the CFT (for a review see [54]). Hawking,
Maldacena, and Strominger[55] applied these ideas to understand the entropy in a
two dimensional de Sitter brane world, embedded in AdS, in terms of entanglement
entropy of the conformal field theory vacuum across the de Sitter horizon, and they made
arguments for the extension of this result to higher dimensions. Somewhat similarly,
Das and Zelnikov[56] have identified the mapping between the Unruh radiation for an
accelerated observer in AdS and a thermal appearance of the ground state in the CFT
for a corresponding class of observers. Presumably the acceleration horizon entropy can
be identified with some definite CFT entropy along these lines, although that has not
been shown explicitly.
In summary, it is fair to say that, wherever the question can be addressed, all
of these approaches to counting black hole states apply in principle equally well to
all Killing horizons. This seems to be consistent with the universal interpretation of
horizon entropy we are advocating here.

6 WHAT DOES HORIZON ENTROPY COUNT?

We have argued that the entire framework of black hole thermodynamics and in par-
ticular the notion of black hole entropy extends to any causal horizon. What are the
implications of this conclusion?
First of all, our attention is deflected away from the black holes and towards the
horizons in black hole thermodynamics. It is sometimes considered a mystery how a
black hole horizon could be capable of carrying so much entropy when after all it has no
local significance in but is rather defined teleologically in terms of the future evolution
of the spacetime. For example, it is regarded as puzzling that when a star collapses
and forms a black hole, the entropy suddenly rockets up to a value many orders of
magnitude greater than it was in the star, “just because” the horizon has formed. This
becomes much less mysterious when it is realized that in essence the black hole really
has nothing to do with it. Any causal horizon is endowed with a surface entropy density
of 1/4.
The realization that horizon entropy is an intrinsically observer dependent notion
raises the obvious question of what are the states that the horizon entropy counts? The
notion that it counts the number of internal configuations, i.e. configurations behind
the horizon, was argued against in [57] on various grounds. It seems only even possibly
viable if the holographic conjecture[58] holds, i.e. if the entire description of the world
behind any horizon can be fully described on its bounding surface. It was argued in
[57] that the holographic conjecture is at best valid when there is no trapped surface
behind the horizon, but it may otherwise in some sense be true. Whether or not it is
true, the fact remains that, for the observers who remain confined to the “outside” of
the horizon, the horizon entropy somehow captures the number of ways that the world
beyond the horizon can affect the world outside.
Black hole pair creation rates are sometimes cited for a counter-argument to this
viewpoint, since the rates are proportional to exp A/4 which is interpreted as a density
of states factor counting degeneracy of all black hole states. However it must be recalled
that the instanton method for computing a tunneling rate just yields the probability
for the most probable transition and the ones in its immediate vicinity. Hence the

16
degeneracy factor is not counting all interior states of the black hole since these are
beyond the scope of the instanton method.
One clue to the nature of the horizon states counted by A/4 comes from an old
analysis by Candelas and Sciama[36, 37]. They showed how the relationship between
near equilibrium transition rates for a system in contact with a horizon and horizon
area is extended to non-equilibrium processes. They interpreted the viscous dissipation
rate of a shearing horizon, via the fluctuation-dissipation theorem, in terms of the
quantum gravitational spectrum of shear fluctuations. This explains “why” a horizon
has a coefficient of viscosity, and suggests that it is, at least in part, the quantum shear
states of the fluctuating horizon that the entropy counts. This seems a good beginning,
but it is surely not the whole story. In addition to the shear viscosity term in Eq. (6)
for the horizon evolution there is also a bulk viscosity (θ2 ) term. Morover, even when
both these terms are absent the horizon acts as a ”perfect dissipator”[36, 37] via just
the area expansion.
Another idea is that the horizon entropy arises from the entanglement of short dis-
tance field fluctuations on either side of the horizon[59]. This notion is quite appealing
since it traces the entropy directly to the defining character of the horizon as a causal
barrier that hides information, and it naturally accounts for the scaling with area.
Moreover, it allows the generalized second law to be understood as a consequence of
causality[60]. However, it is problematic in the quantitative measure of missing informa-
tion (which appears to be infinite in ordinary field theory and to depend on the number
of species) and in the neglect of the quantum fluctuations of the horizon itself. These is-
sues are tied up with the renormalization of Newton’s constant[61], though a completely
satisfactory understanding at the statistical level is far from being at hand. The finite-
ness problem is absent in the loop quantum gravity approach[52], which can be viewed
as a counting of intertwiners that characterize the possible ways a spin network inside
the black hole could link to the one in the exterior. However, that approach remains
to be tied to the semiclassical limit in a quantitatively convincing manner. Interesting
perspectives on some of these issues are given in the review article by Bekenstein[16].
While no definitive answer as to the ultimate nature of horizon entropy seems im-
mediately at hand, an abundance of insight has been gleaned from the three decades of
work. Perhaps the time is ripe to synthesize this insight and make the leap to a new
conception.

ACKNOWLEDGMENTS
We thank A. Ashtekar, P. Chruściel, and G. Galloway for helpful correspondence.
This work was supported in part by the National Science Foundation under grant No.
PHY98-00967 and NATO Collaborative Linkage Grant CLG.976 417.

References
[1] D. Christodoulou, “Reversible and irreversible transformations in black-hole
physics,” Phys. Rev. Lett. 25 1596 (1970).
[2] R. Penrose and R. M. Floyd, “Extraction of rotational energy from a black hole,”
Nature (Physical Science) 229, 177 (1971).
[3] S. W. Hawking, “Gravitational radiation from colliding black holes,” Phys. Rev.
Lett. 26, 1344-1346 (1971).
[4] J. D. Bekenstein, “Black holes and entropy,” Phys. Rev. D 7, 2333 (1973).

17
 [5] J. D. Bekenstein, “Generalized second law of thermodynamics in black-hole
physics,” Phys. Rev. D 9, 3292 (1974).
[6] S. W. Hawking, “Particle creation by black holes,” Commun. Math. Phys. 43, 199
(1975).
[7] J. M. Bardeen, B. Carter, and S. W. Hawking, “The four laws of black hole me-
chanics,” Commun. Math. Phys. 31, 161 (1973).
[8] J. D. Bekenstein, “Statistical black-hole thermodynamics,” Phys. Rev. D 12, 3077
(1975).
[9] S. W. Hawking, “Black hole and thermodynamics,” Phys. Rev. D 13, 191 (1976).
[10] W.H. Zurek and K.S. Thorne, “Statistical mechanical origin of the entropy of a
rotating, charged black hole” Phys. Rev. Lett. 54 2171 (1985).
[11] W. G. Unruh and R. M. Wald, “Acceleration radiation and generalized second law
of thermodynamics,” Phys. Rev. D 25, 942 (1982); “Entropy bounds, accelera-
tion radiation, and the generalized second law,” Phys. Rev. D 27, 2271 (1983);
M. J. Radzikowski and W. G. Unruh, “Correction to ‘Acceleration radiation and
the generalized second law of thermodynamics’,” Phys. Rev. D 37, 3059 (1988).
[12] V. P. Frolov and D. N. Page, “Proof of the generalized second law for quasistation-
ary semiclassical black holes,” Phys. Rev. Lett. 71, 3902 (1993).
[13] G. W. Gibbons and S. W. Hawking, “Cosmological event horizons, thermodynam-
ics, and particle creation,” Phys. Rev. D 15, 2738 (1977).
[14] G. W. Gibbons and S. W. Hawking, “Action integrals and partition functions in
quantum gravity,” Phys. Rev. D 15, 2752 (1977).
[15] J. D. Bekenstein, “Extraction of energy and charge from a black hole,” Phys. Rev.
D 7, 949 (1973).
[16] “Do we understand black hole entropy?”, in Proceedings of the Seventh Marcel
Grossmann Meeting, Robert T. Jantzen, G. Mac Keiser, and Remo Ruffini, eds.
(World Scientific, 1996).
[17] T. Jacobson, “Thermodynamics of spacetime: the Einstein equation of state,”
Phys. Rev. Lett. 75, 1260 (1995).
[18] P. Martinetti and C. Rovelli, “Diamonds’s temperature: Unruh effect for bounded
trajectories and thermal time hypothesis,” arXiv:gr-qc/0212074.
[19] I. Racz and R. M. Wald, “Extensions of spacetimes with Killing horizons,” Class.
Quantum Grav. 9, 2643 (1992).
[20] B. Carter, “Black hole equilibrium states,” in Black holes, C. DeWitt and B.S.
Dewitt, eds. (Gordon & Breach, New York, 1973).
[21] I. Racz and R. M. Wald, “Global extensions of spacetimes describing asymptotic
final states of black holes,” Class. Quantum Grav. 13, 2643 (1996).
[22] R. Penrose, “Gravitational collapse,” in Graviational Radiation and Gravitational
Collapse, C. DeWitt-Morette, ed. (Reidel, Dordrecht, 1974); “Singularities of
spacetime,” in Theoretical Principles in Astrophysics and Relativity, N.R. Lebovitz,
W.H. Reid, and P.O. Vandervoort, eds. (The Univ. of Chicago Press, 1978).
[23] S.W. Hawking and G.F.R. Ellis, The Large Scale Structure of Space-Time (Cam-
bridge University Press, 1973).
[24] R.M. Wald, General Relativity (The University of Chicago Press, 1984).

18
[25] P.T. Chruściel, E. Delay, G.J. Galloway, and R. Howard, “Regularity of horizons
and the area theorem,” Annales Henri Poincaré 2, 109 (2001).
[26] P. C. Davies, “Cosmological Horizons And The generalized second law of ther-
modynamics,” Class. Quant. Grav. 4, L225 (1987); “Cosmological Horizons And
Entropy,” Class. Quant. Grav. 5, 1349 (1988).
[27] S. A. Hayward, T. Shiromizu, and K. Nakao, “A cosmological constant limits the
size of black holes,” Phys. Rev. D 49, 5080 (1994).
[28] T. Shiromizu, K. Nakao, H. Kodama, and K. Maeda, “Can large black holes collide
in de Sitter space-time? An inflationary scenario of an inhomogeneous universe,”
Phys. Rev. D 47, 3099 (1993).
[29] K. Maeda, T. Koike, M. Narita and A. Ishibashi, “Upper bound for entropy in
asymptotically de Sitter space-time,” Phys. Rev. D 57, 3503 (1998).
[30] R.M. Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermo-
dynamics, (University of Chicago Press, 1994).
[31] S.W. Hawking and J.B. Hartle, “Energy and angular momentum flow into a black
hole,” Comm. Math. Phys. 27, 283 (1972).
[32] B. Carter, “The general theory of mechanical, electromagnetic and thermodynamic
properties of black holes,” in General Relativity: An Einstein Centenary Survey,
S.W. Hawking and W. Israel, eds. (Cambridge University Press, 1979).
[33] S.W. Hawking, G.T. Horowitz, and S.F. Ross, “Entropy, area, and black hole
pairs,” Phys. Rev. D 51, 4302 (1995).
[34] J. Pfautsch, cited in P.C.W. Davies, “Mining the Universe,” Phys. Rev. D 30 737
(1984); see also H. Narnhofer, I. Peter, and W. Thirring, W., “How hot is the
de Sitter space?,” Int. J. Mod. Phys. B 10 1507 (1996); S. Deser, and O. Levin,
“Accelerated detectors and temperature in (anti-) de Sitter spaces,” Class. Quan-
tum Grav. 14 L163 (1997); T. Jacobson, “Comment on accelerated detectors and
temperature in (anti-) de Sitter spaces,” Class. Quantum Grav. 15 251 (1997).
[35] Black Holes: The Membrane Paradigm, R. Price, K. Thorne, D.A. MacDonald,
eds. (Yale University Press, 1986), Section VII.E.1
[36] P. Candelas and D.W. Sciama, “Irreversible thermodynamics of black holes,” Phys.
Rev. Lett. 38, 1372 (1977).
[37] D.W. Sciama, “Black holes and fluctuations of quantum particles: an Einstein syn-
thesis,” in Relativity, Quanta, and Cosmology in the Development of the Scientific
Thought of Albert Einstein, vol. II, M. Pantaleo, dir. and F. De Finis, ed. (Giunti
Barbèra, Firenze, 1979); reprinted: (Johnson Reprint Corporation, Harcourt Brace
Jovanovich subsidiary, 1979).
[38] D. Garfinkle, S.B. Giddings, and A. Strominger, “Entropy in black hole pair pro-
duction,” Phys. Rev. D 49, 958 (1994).
[39] S. Massar and R. Parentani, “Schwinger mechanism, Unruh effect and production
of accelerated black holes,” Phys. Rev. D 55, 3603 (1997).
[40] R.B. Mann and S.F. Ross, “Cosmological production of charged black hole pairs,”
Phys. Rev. D 52 2254 (1995).
[41] S.W. Hawking and G.T. Horowitz, “The gravitational Hamiltonian, action, entropy
and surface terms,”Class. Quant. Grav. 13, 1487 (1996).
[42] S. Massar and R. Parentani, “How the change in horizon area drives black hole
evaporation,” Nucl.Phys. B575, 333 (2000); gr-qc/9903027.

19
[43] J.B. Hartle and S.W. Hawking, “Path-integral derivation of black-hole radiance,”
Phys. Rev. 13, 2188 (1976).
[44] W.G. Unruh, “Notes on black hole evaporation”, Phys. Rev. D14, 870 (1976).
[45] R. Parentani, “The recoils of the accelerated detector and the decoherence of its
fluxes,” Nucl. Phys. B454, 227 (1995).
[46] S. Carlip and C. Teitelboim, “The off-shell black hole,” Class. Quant. Grav.
12,1699 (1995).
[47] P. Kraus and F. Wilczek, “Self-interaction correction to black hole radiance,” Nucl.
Phys. B433, 403 (1995).
[48] E. Keski-Vakkuri and P. Kraus, “Microcanonical D-branes and back reaction,”
Nucl. Phys. B491, 249 (1997).
[49] R. Parentani, “Time-dependent perturbation theory in quantum cosmology,”
Nucl.Phys. B492, 501 (1997).
[50] R. Brout et al., “A primer for black hole quantum physics,” Phys. Rep. 260, 330
(1995).
[51] S. Carlip, “Entropy from conformal field theory at Killing horizons,” Class. Quant.
Grav. 16, 3327 (1999); “Near-horizon conformal symmetry and black hole entropy,”
Phys. Rev. Lett. 88, 241301 (2002).
[52] C. Rovelli, “Black hole entropy from loop quantum gravity,” Phys. Rev. Lett. 77,
3288 (1996); K. V. Krasnov, “Geometrical entropy from loop quantum gravity,”
Phys. Rev. D 55, 3505 (1997); A. Ashtekar, J. Baez, A. Corichi, and K. Krasnov,
“Quantum geometry and black hole entropy,” Phys. Rev. Lett. 80, 904 (1998);
A. Ashtekar, J. C. Baez and K. Krasnov,“Quantum geometry of isolated horizons
and black hole entropy,”Adv. Theor. Math. Phys. 4, 1 (2000).
[53] G.T. Horowitz, “Quantum states of black holes,” in Black holes and relativistic
stars, R.M.. Wald, ed. (The University of Chicago Press, 1998).
[54] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, “Large N field
theories, string theory and gravity,” Phys. Rept. 323, 183 (2000).
[55] S. Hawking, J. Maldacena, and A. Strominger, “DeSitter entropy, quantum entan-
glement and AdS/CFT,” JHEP 0105, 001 (2001).
[56] S. R. Das and A. Zelnikov, “Unruh radiation, holography and boundary cosmol-
ogy,” Phys. Rev. D 64 104001 (2001).
[57] T. Jacobson, “On the nature of black hole entropy,” in General Relativity and Rel-
ativistic Astrophysics: Eighth Canadian Conference, AIP Conference Proceedings
493, C. Burgess and R.C. Myers, eds. (AIP Press, 1999), pp. 85-97.
[58] R. Bousso,“The holographic principle,” Rev. Mod. Phys. 74, 825 (2002).
[59] L. Bombelli, R. K. Koul, J. H. Lee, and R. D. Sorkin, “A quantum source of entropy
for black holes,” Phys. Rev. D 34, 373 (1986).
[60] R. D. Sorkin, “Toward a proof of entropy increase in the presence of quantum
black holes,” Phys. Rev. Lett. 56, 1885 (1986); “The statistical mechanics of black
hole thermodynamics,” in Black holes and relativistic stars, R.M. Wald, ed. (The
University of Chicago Press, 1998).
[61] L. Susskind and J. Uglum, “Black hole entropy in canonical quantum gravity and
superstring theory,” Phys. Rev. D 50, 2700 (1994). C. Holzhey, F. Larsen and
F. Wilczek, “Geometric and renormalized entropy in conformal field theory,” Nucl.

20
Phys. B 424, 443 (1994). T. Jacobson, “Black hole entropy and induced gravity,”
arXiv:gr-qc/9404039. F. Larsen and F. Wilczek, “Renormalization of black hole
entropy and of the gravitational coupling constant,” Nucl. Phys. B 458, 249 (1996).

21